BJFORE Procedure

@BJFORE computes forecasts using Box-Jenkins techniques.

@BJFORE(options) series forestart foreend forecasts

Parameters

series	(input) series to be forecast
forestart, foreend	range of entries to forecast
forecasts	(output) series for computed forecasts

Options

The following sets of options choose what transformations are applied to the data before estimating the ARMA model. The forecasts are initially done to the stationary data that results and then backtransformed to give the forecasts of the input series.

DIFFS=Number of regular differences [0]

SDIFFS=Number of seasonal differences [0]

[CONST]/NOCONST

Includes a constant term in the model.

TRANS=[NONE]/LOG/ROOT

Transformation to apply to data

These pick the model to apply. You have to identify this first.

ARS=Number of autoregressives[0]

MAS=Number of moving averages[0]

SAR=Number of seasonal autoregressives[0]

SMA=Number of seasonal moving averages[0]

SPAN=Seasonal span [CALENDAR seasonal]

[PRINT]/NOPRINT

Use NOPRINT to turn off display of BOXJENK estimation output

BJTRACE/[NOBJTRACE]

Turns on the TRACE option on the BOXJENK instruction, which displays a trace of the estimation process.

RSTART=start of estimation range [earliest possible}

REND=end of estimation range [forestart-1]

These set the range of entries used to estimate the mode. RSTART defaults to the earliest possible entry. REND defaults to (forestart-1).

Example

This estimates an ARIMA model with both regular and seasonal differences, and 1 AR and 1 seasonal MA, on the log of M1NSA (non-seasonally adjusted M1). The forecasts are computed over 2008:3 (written here as 2008:2+1 to make it clearer what the end of the estimate range is) to 2009:2. @BJFORE itself does not create any output, so we show the graph.

@bjfore(ars=1,smas=1,trans=log,diffs=1,sdiffs=1,constant) m1nsa 2008:2+1 2008:2+12 m1fore m1resids

graph(header="M1NSA Out of Sample Forecasts and Actuals") 2

# m1nsa 2000:1 *

# m1fore